rann 0.2.1 → 0.2.2

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA1:
-  metadata.gz: 6dff2d631ccf502189d84d05fa9835cb8fa68c9d
-  data.tar.gz: cb1df08ddd1421e2c4a4f736fb8d946bafabbd8e
+  metadata.gz: 1d6d82c495eff427af738f4734993fd2611848b9
+  data.tar.gz: de50063a5696ae822b565ce61629f4474e237d78
 SHA512:
-  metadata.gz: aaa784026bb0e1a7efd2234b1bb906aee50dfb392677825264f8b77fd5f9a23a04679a6c23306d7a70bc8d0e15a553d804e76afdd4a448198d9f67b05ad50392
-  data.tar.gz: ee77695cfb639fa917a5684628ba9681c4484a60c863b814805697f1fe81f9da834f866069136e38b60afcc1322575627714716bd228764b7a5ddd978860c9f5
+  metadata.gz: 478e37a62c101eec3f8bc23125c308fb49f27257b80a119dea4747c481fe65f5d53397db7013cf4642e097ff81699471d4e6ff860720e2366fd90cf2760ca61a
+  data.tar.gz: 1a131eb754900d36f2353064cd8b05f1e179adbbee95e25063511a029636378490dff0b2f1bc7678a31027d0dbc6b0fee6bd20df14ef25ace465b6b31fff828e

data/lib/rann/backprop.rb

@@ -34,9 +34,12 @@ module RANN
       avg_gradients   = Hash.new{ |h, k| h[k] = 0 }
       avg_batch_error = 0
 
-      # force longer bits of work per iteration, to maximise CPU usage
-      # less marshalling data etc, more work.
-      grouped_inputs  = in_groups(inputs, [1, opts[:processes]].max * 10, false).reject &:empty?
+      # force longer bits of work per iteration, to maximise CPU usage less
+      # marshalling data and process overhead etc. best for small networks. for
+      # larger networks where one unit of work takes a long time, and the work
+      # can vary in time taken, use num_groups == inputs.size
+      num_groups     = opts[:num_groups] || ([1, opts[:processes]].max * 10)
+      grouped_inputs = in_groups(inputs, num_groups, false).reject &:empty?
       reduce_proc =
         lambda do |_, _, result|
           group_avg_gradients, group_avg_error = result
@@ -110,8 +113,8 @@ module RANN
       error = mse targets, outputs
 
       # backward pass with unravelling for recurrent networks
-      node_deltas = Hash.new{ |h, k| h[k] = Hash.new(0.to_d) }
-      gradients = Hash.new(0)
+      node_deltas = Hash.new{ |h, k| h[k] = Hash.new 0.to_d }
+      gradients = Hash.new 0.to_d
 
       initial_timestep = inputs.size - 1
       neuron_stack = network.output_neurons.map{ |n| [n, initial_timestep] }

data/lib/rann/version.rb

@@ -1,3 +1,3 @@
 module RANN
-  VERSION = "0.2.1"
+  VERSION = "0.2.2"
 end

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: rann
 version: !ruby/object:Gem::Version
-  version: 0.2.1
+  version: 0.2.2
 platform: ruby
 authors:
 - Michael Campbell
 autorequire: 
 bindir: exe
 cert_chain: []
-date: 2017-11-24 00:00:00.000000000 Z
+date: 2017-11-26 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: parallel